RoboGraphBench - Indoor: leaderboard
RoboGraphBench (Wang and Li, 2026) compiles long-horizon robot tasks into symbolic scene graphs whose task-relevant state must be tracked through exploration, action failures and interventions, and runs models as closed-loop high-level planners. This board covers 189 navigation-enabled indoor episodes from 27 scenes, each in a baseline and six intervention conditions. Task success rate in percent; higher is better.
Metric: Task success rate (%; episodes that reach the goal and stop correctly, averaged over the baseline and six intervention conditions; semantic closed-loop evaluation on partially observed symbolic scene graphs, at most 100 steps; 189 navigation-enabled indoor episodes from 27 scenes). Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 82.5 | #48 |
| 2 | Claude Opus 4.7 | 80.4 | #75 |
| 3 | GLM-5.2 | 57.1 | #137 |
| 4 | Gemini 3.1 Pro (Preview) | 42.3 | #91 |
| 5 | GPT-5.4 | 41.8 | #121 |
| 6 | Qwen 3.6 Plus | 28.6 | #217 |
| 7 | DeepSeek V4 Pro | 23.3 | #181 |
| 8 | Qwen 3.7 Plus | 21.7 | #203 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=robographbench-indoor · How It Works · Data refreshed daily, snapshot 2026-11-12.